cemantle

Collaborate

You can collaborate as a team on a game by generating a code to share. Users of this code will receive the words you find in the top 1000, and you’ll receive theirs. The code is good for the day.
This feature uses some resources in limited quantity, therefore:

There’s no point creating a team right before midnight, it will be lost when the time comes.
Please do not share the code publicly on social networks!
Thanks for playing the game 🙂

Create a team:

Join a team:

You’re in team

You can continue your game on another device by identifying yourself with Twitter. You just have to authenticate with Twitter on both devices for the game to be shared. No data from your Twitter account other than your ID is used.

Authenticate with Twitter:

You’re signed in as

History

You can synchronize your history across devices by identifying yourself with Twitter. No data from your Twitter account other than your ID is used.

Authenticate with Twitter:

You’re signed in as .

Synced . 🔃 Sync

#	Word	Guesses	🥳	Solvers

Share

Colors

🎨 Theme	🔆 Mode
Mardi Gras	Light
Halloween	Dark
	System

Find the secret word!

The objective of the game is to find the secret word by trying to get as close as possible contextually. Each word is assigned a temperature whose values are given in the scale on the left. If your word is in the nearest 1000 words, a progress bar graduated from 1 to 1000‰ will appear.

The proximity of a word is not orthographic (like this other game) but semantic or contextual. It’s evaluated not using a dictionary, but a database of texts of more than three billion words in which a relative “distance” between each word has been calculated. Two close words in such a lexical field are not necessarily synonyms. For example, it may be that an adjective and its opposite are considered close because they can qualify the same thing.

Uppercase letters are ignored and proper nouns are generally not allowed. You’re looking for a singular noun, adjective, unconjugated verb or adverb.

You will need more than 6 guesses; probably dozens of them. The ranking given to you at the end of the game is your position in the list of players who have found the word of the day, it’s independent of the number of guesses.

There’s a new random word every day at midnight US Pacific Time, or local time.

Uncover the secret Wikipedia page!

The objective of the game is to find the secret Wikipedia page by revealing in successive attempts the words composing its introduction.

Correct words will appear in clear as you guess them. Words that are close enough will stay grayed out with a level of gray proportional to their proximity to the correct word in the lexical field. This calculation is similar to the one used in cemantle.

When words forming the Wikipedia page title are uncovered, you win! Please note that the title words are either correct or not, they're never grayed out. The root of a word could be enough to reveal its plural and conjugated forms. Uppercase letters aren't necessary.

You will need more than 6 guesses; probably dozens of them. The ranking given to you at the end of the game is your position in the list of players who have found the page of the day, it’s independent of the number of guesses.

There’s a new random page every day at noon US Pacific Time, or local time.

FAQ

How does the algorithm behind cemantle work?

Imagine you’re sent to a desert island with only a book for your entertainment, and that book is written in a language you don’t know. Let’s say Hawaiian (if you know that language, pick another one). Upon your return, you’re asked to summarize the story you read. You’ll have no idea: the book contains no images and nothing can make you understand the meaning of the words. There is no Rosetta stone on the island! All you can tell is that the book contains words: sets of letters separated by spaces.
However, you’d be surprised to realize that you can answer a few questions regarding the language. For example, if you’re asked what word would go well with kumulāʻau, you would say hua. If you’re asked what could replace manu in a sentence you could say holoholona. Thus, without even knowing the meaning of those words, you can associate them, and there’s a good chance your friend would be satisfied with your answers. You have simply observed the frequency of certain sequences of words as well as the position of those words in the sequences, and can therefore infer associations with a certain degree of confidence.
What the algorithm does behind cemantle is exactly that: it doesn’t know English, it has no dictionary or grammar book to understand a text, a sentence or even a word. It doesn’t know what a noun, a verb or an adjective is, nor a synonym or antonym. All it has is a large enough corpus of text to form the word associations with a good chance of being statistically correct. While it can give results that seem illogical to a human, the data comes from existing texts and there’s always a reason why the association was made, even if it doesn’t seem obvious at first glance.
How are associations made?

Here is an example:
- Alice walks her little dog.
- Bob feeds his big dog.
If those sentences are repeated a certain number of times in a text, we can naturally conclude that "little and dog" and "big and dog" are associated because they’re physically close in the sentence. But also "little and big" because even though they’re not physically close (they don’t appear in the same sentence), they are interchangeable. The proximity of the word dog creates an association between them although they mean the opposite. On the other hand, we’ll never see the phrase “Charlotte throws the ball to her canine dog”, meaning that dog and canine are not close, at least rarely physically. Only interchangeability could fix that (e.g. David throws the ball to his canine companion). Following the same principle, "walks and feeds" must be associated, which may seem surprising, as well as "Alice and Bob", but after all, maybe they are 😊. It’s important to remember that it’s all about statistics, the frequency of those associations in the texts contributes to the temperature of each word.
How are temperatures calculated?

It was in 2013 that a team of Google engineers had the idea of representing the words of a text in a multi-dimensional space (we are talking about hundreds of dimensions here) by following the rules of association described above and considering their relative position to other words. Each word is assigned a vector in each dimension of this space, thus constituting a coordinate system. This model is known as word2vec. Once that’s done, it’s easy to calculate the "distance" between any pair of words. This distance is the temperature displayed in cemantle.
How is the choice of words made?

The list of words close to the secret word is entirely determined by the algorithm, there is no human intervention. The model used is certainly not “perfect” and results can always hold some surprises for players, but it uses one of the most complete corpora available which should minimize such annoyances.
The choice of the secret word is random. Secret words are all relatively common words in the English language, everyone should know them. If a word relates to current events, if it’s similar to another word of the day, if it may sound offensive or biased, if it seems too easy or too difficult to find, it’s merely a coincidence.

Announcement

Words close to intelligent

#	Word	°C		‰

‰	🌡	°C
1000	🥳	100
999	😱	58.33
990	🔥	46.88
900	🥵	29.79
1	😎	17.97
	🥶	0
	🧊	-100

. ( )